The new JChem Microservices uses the latest generation search technology that can handle large datasets. Microservice architecture has a modular setup, so instead of one monolithic application, we have smaller modules with specific functionalities. JChem Microservices provide small, separate modules for different areas of ChemAxon functionalities like chemical dataset searching, conversion between chemical file formats, or chemical property calculation. This is scalable, easily manageable, and cloud-agnostic. In this webcast, we will show you how to set up a highly available architecture using Microservices, and demonstrate an example using Enamine search as a service.

1. JChem Microservices
Andras Volford

2. Chemical toolkit
Modular Scalable Easily
manageable
Cloud
agnostic

3. What Microservices are good for
Set of Chemistry Tools to assist drug
discovery.
Extend the functionality of your idea with
Chemistry.

4. Modules

5. Chemaxon specific modules
Structure Manipulation
Demo site
Structure Checker
Calculations
DB
IO
Reactor
Markush

6. The Three Little Pigs
● Gateway:
Route to APIs, also support predicates and filters
● Discovery:
Instances can be registered and clients can discover
the instances
● Config:
Resource-based API for external configuration
Not ChemAxon specific
Config server Discovery
Gateway
Chemical
intelligence
User frontier
Rest call

7. Database module

8. Database module
Store and search chemical structures and reactions.
Duplicate
Full Fragment
Substructure
Similarity

9. Hit As You Draw
First 100 hits return quickly.
Hits ordered by relevance to the query structure.
Demo

10. 2019 - Monolith setup
Searching on Enamine REAL

11. Goal
Use Enamine REAL molecule set
containing 720 million structures
for substructure
and similarity search

12. Setup
JChem Microservices DB is used for
chemical search
Memory requirement: 70577 MB RAM
AWS EC2 r5.4xlarge
16 vCPU 128 GB RAM

13. Convert SMILES csv 2 json
Enamine_REAL_database_smiles.zip 19GB
parallel conversion using 1M chunks finally ~90GB (finished in a day)
Import json
Took 3d 9h
Start service
Cache load ~ 5 min
Import data

14. Query structures

15. Query structures

16. Query structures

17. Query structures

18. Substructure search

19. Substructure search

20. Substructure search
42

21. Substructure search
13
42

22. Substructure search

23. Similarity search
5
8

24. Similarity search

25. Distributed setup
Searching on Enamine REAL

26. Monolithic vs sharded setup

27. Andras Volford
avolford@chemaxon.com

28. Finding leads and analogs in
Enamine REAL

29. • finding purchasable compounds to prioritize scaffold ideas
• available virtual catalogs
– Enamine REAL at 1010
– Wuxi Galaxi at 109
– many pharma companies launching similar size prop. spaces
• JChem Microservices offers scalable platform
• JChem offers all typical search methods and query fine-tuning
options
Exploiting the possibilities in virtual commercial spaces

30. similarity searches
total hits at >0.7 similarity score: 346567
reference
https://pubmed.ncbi.nlm.nih.gov/27983835

31. total hits at >0.7 similarity score: 1229783
similarity searches
reference

32. similarity searches
total hits at >0.7 similarity score: 2118692
reference

33. substructure searches
total matches: 17226
reference

34. substructure searches
ordered by similarity to query - search time reduced
reference

35. • decoration free query structures and similarity search (tanimoto
>0.7) capture large pool of options
• some substructures match large sets of compounds
• ordered retrieval is beneficial
• fast retrieval is essential
Search options

37. • 50 benchmark compounds from IRAK4 and Pan-Janus related
literature
• substructure searches ordered by similarity to query
• similarity search cutoff >0.7 Tanimoto
Enamine REAL search performance in Design Hub
max hits substructure similarity
200 3.36s 3.79s
2000 3.53s 3.87s
20000 4.30s 4.27s

39. performance

43. curl -X GET https://cognito-idp.eu-west-1.amazonaws.com/<redacted>/.well-known/openid-configuration | jq '.token_endpoint'
curl -X POST --user $CLIENT_ID:$CLIENT_SECRET -d "grant_type=client_credentials" $TOKEN_URL | jq '.access_token'
curl -X POST https://<redacted>.design-analysis.cxcloud.io/runSearch -H 'Content-Type: application/json'
-H "Authorization: Bearer $BEARER_TOKEN"
-d '{ "molecule": "CC1CS(=O)(=O)CCN1N=CC1=CC=C(O1)N(=O)=O", "searchType": "similarity", "threshold": "0.7", "limit": "100000", "searchTimeout": "30000" }'
[{“id":"Z2718472587","molecule":"O=C(NN1CCS(=O)(=O)CC1)C1=CC=C([N+](=O)[O-])O1","similarity":0.812},{"id":"Z2720224253","molecule":"O=[N+]([O-
])C1=CC=C(C=NNC2CCCS(=O)(=O)C2)O1",...]
API usage

44. Design Hub
hypothesis capturing
compound design and prioritization
progression tracking with CROs

46. Andras Stracz
astracz@chemaxon.com